Bump required Python 3.7 -> 3.8, drop TF1 support

[NeoLegends]

Closes #1326

Opening this to run CI and such.

[NeoLegends]

By dropping support for 3.7 we also drop support for TF1.

It is my understanding that some folks might still be using TF1.x for some of their experiments. However, these experiments are probably also using ancient RETURNN versions, so I believe it's ok to drop support.

@albertz WDYT?

[albertz]

By dropping support for 3.7 we also drop support for TF1.

Why?

[NeoLegends]

Why?

TF1 never supported Python > 3.7. See the CI log. There are no pip packages available for TF1 for 3.8 and onwards.

[albertz]

Please check with @curufinwe if dropping TF1 is ok.

[albertz]

Suggested change

	pip dependencies are listed in ``requirements.txt`` and ``requirements-dev``, although some parts of the code may require additional dependencies (e.g. ``librosa``, ``resampy``) on-demand.
	Pip dependencies are listed in ``requirements.txt`` and ``requirements-dev``,
	although some parts of the code may require additional dependencies (e.g. ``librosa``, ``resampy``) on-demand.
	For TensorFlow-based setups, RETURNN requires TF >=2.
	For PyTorch-based setups, RETURNN requires Torch >=1.

(I actually don't really know our min required Torch version... But this here would be some good starting point.)

[albertz]

In the failing test, can you filter out [IWE] tensorflow/ from stdout? I think that should fix the test.

[albertz]

Ah, in the failing test checking the stdout, the SyntaxWarnings are also new. Let's either fix the syntax warning or filter those out as well. Fixing the SyntaxWarning, there is actually a separate PR #954 for that, maybe let's continue that (it's not totally trivial).